MarketPulse

AI-powered market insights and predictive analytics for the modern investor

Comment

MarketPulse: AI-Powered Market Intelligence

Inspiration

As retail investors ourselves, we often found that staying on top of market news and understanding its implications was overwhelming. Professional traders have teams of analysts and expensive tools, but individual investors are left to interpret complex financial data on their own. When a breaking news story drops, understanding how it might affect your specific portfolio can be challenging.

We wanted to create a tool that democratizes financial intelligence and levels the playing field between Wall Street professionals and everyday investors. The question that drove us was simple: "What if we could build an AI assistant that understands your investment interests, tracks your holdings, and explains financial news in a personalized way?"

What it does

MarketPulse is a comprehensive financial intelligence platform that:

  • Tracks your portfolio: Users can select stocks they own and areas of interest (like Renewable Energy or AI) for personalized insights
  • Analyzes news sentiment: Our algorithm scrapes financial news and evaluates potential market impact
  • Provides AI-powered assistance: The built-in AI assistant answers questions about your holdings, recommends stocks based on your interests, and explains complex financial reports
  • Offers RAG model insights: We've implemented a Retrieval-Augmented Generation model trained on company 10-K reports to provide deep insights into specific companies

The platform goes beyond simple stock tracking by interpreting news events and explaining their potential impact on your specific holdings. It bridges the gap between raw data and actionable intelligence.

How we built it

We developed MarketPulse using a modern tech stack:

  • Frontend: JavaScript using React for a responsive, intuitive interface
  • Backend: Python with Flask for our API and web scraping infrastructure
  • AI Components:
    • Google Gemini API for the conversational AI assistant
    • Stock trend and magnitude prediction model trained on financial news and technical data
    • RAG implementation using LangChain to process and query 10-K reports
  • Authentication: Auth0 for secure user management
  • Data Sources: Yahoo Finance, multiple news websites, and SEC annual 10-K reports

The development process began with building a robust news scraping and analysis pipeline. We then developed the trend prediction model, trained on thousands of financial news articles matched with subsequent market movements. The RAG model was implemented last, using vector embeddings to make company reports queryable in natural language.

Challenges we ran into

Building MarketPulse wasn't without its hurdles:

  1. Data preparation: Collecting and preparing the train data for the trend prediction model was very challenging. It included - web scraping historical financial news article, identifying stock names in the text, and calculating historical technical stock data.

  2. Prediction model accuracy: Early versions of our prediction model was too primitive, often misinterpreting nuanced financial news. We had to retrain the model and perform hyperparameter tuning multiple times with more sophisticated and relevant features.

  3. RAG implementation: Training the RAG model on dense financial documents like 10-K reports required significant optimization to maintain both speed and accuracy.

  4. Real-time performance: Processing news and generating insights quickly enough to be actionable required careful optimization of our backend systems.

  5. UI/UX design: Creating an interface that presents complex financial data in an approachable way took multiple iterations and user testing sessions.

Accomplishments that we're proud of

Despite the challenges, we're proud of several achievements:

  • Successfully implementing a RAG model that can answer specific questions about company reports in natural language
  • Creating an AI assistant that provides genuinely useful financial insights rather than generic advice
  • Developing a trend prediction model that can interpret complex financial news with MSE of 0.0006
  • Building a clean, intuitive interface that makes sophisticated financial tools accessible
  • Integrating multiple data sources to provide a comprehensive view of market conditions

What we learned

This project taught us valuable lessons about:

  • The importance of data quality in machine learning projects
  • Techniques for optimizing large language models for specific domains
  • Methods for extracting meaningful signals from noisy financial news
  • Approaches to presenting complex data in accessible ways
  • The challenges of building systems that need to process and respond to real-time events

Most importantly, we learned that building AI tools for finance requires a careful balance between automation and human judgment. Our system works best when it augments human decision-making rather than trying to replace it.

What's next for MarketPulse

We have ambitious plans for MarketPulse's future:

  • Expanded coverage: Include more asset classes beyond stocks (bonds, commodities, cryptocurrencies)
  • Portfolio optimization: Add AI-powered suggestions for portfolio diversification and risk management
  • Advanced alerts: Develop smarter notification systems that alert users only to truly significant events
  • Community features: Build tools for sharing insights and collaborating with other investors
  • Mobile application: Create a native mobile experience for on-the-go market intelligence

Our vision is to continue bridging the gap between professional and retail investors by making sophisticated financial intelligence accessible to everyone. We believe that better-informed investors make for healthier markets, and MarketPulse is our contribution to that goal.

Built With

Share this project:

Updates